AMD Execution Thread [2023]

Status
Not open for further replies.
Whoa, those API overheads?!! I assume lower is better? How did they drop those that much, it's awesome.
They count draw calls I believe so higher is better. But that test has been deprecated for some time so no way to tell if these results even mean anything on modern h/w.
 
has there been any reliable leaks on what AMD is doing with RDNA 3 ? Are they just going with more CU's for ray tracing or something else ?
 
Seems AMD is not planning a new release soon, they are not going to announce anything consumer GPU related at Computex, and they already released N33 before N32, which is unheard of. One of the comments on the matter caught my attention.

Seems AMD is estimating that demand just isn't there, the numbers thrown around from both the 4060Ti and 7600 launches are 90K globally each, which is obscenely low. Not only are dGPU sales down overall, they're more down for AMD which last we saw any data on market share, it was sitting at somewhere around 9-10% (a drop from 18% just a couple quarters prior).
There is already too much RDNA2 inventory out there that is competing directly with RDNA3. Let's face it, RDNA3 doesn't bring any new features that it can latch on to to make it more appealing than prior gen cards. AV1? most people don't even know what AV1 encoding is. Slightly better RT? If someone's buying a card for RT, they'll get Nvidia, so on and so on.
Lastly, AMD is in a tight bind with segmenting the 7700 and 7800. Where do they slot it and at what price point? 6950XT is sub $600 and 7900XT is already reaching the $750 mark in some markets. Where do you slot in a 7700 and 7800 without it interfering with the rest of line up? You're either going to kill off your remaining inventory or drive everything else down with it. They're in a bind because the performance just isn't there either to slot in N32 cards or start cutting down N31 and ask the price they'd want to ask for them AND then sell them.
Over the last few years, AMD has been stepping away from the dGPU segment and focusing their attention on bigger markets (money makers) in the enterprise and the custom solutions segment which to be honest, both are killing it in both enterprise and console/handheld solutions. So why would they dump money, resources, and time into couple of midrange cards for the 2-3% of buyers in that market?
 
Seems AMD is not planning a new release soon, they are not going to announce anything consumer GPU related at Computex, and they already released N33 before N32, which is unheard of. One of the comments on the matter caught my attention.
Also doesn't the three Instinct MI300s count as upcoming releases? Not a graphics card, but it is a freaking monster that should pull in a ton of yummy AI cash bubble fun!
 
Also doesn't the three Instinct MI300s count as upcoming releases? Not a graphics card, but it is a freaking monster that should pull in a ton of yummy AI cash bubble fun!

For AI the trouble for AMD is that the software ecosystem is as expensive if not moreso than the Hardware. NV has invested heavily in their software stack focusing on AI, so it's a difficult sell for AMD to gain mass adoption there.

Also, I think the "not planning a new release soon" is WRT consumer GPUs and not data center GPUs (MI300). :)

Regards,
SB
 
For AI the trouble for AMD is that the software ecosystem is as expensive if not moreso than the Hardware. NV has invested heavily in their software stack focusing on AI, so it's a difficult sell for AMD to gain mass adoption there.

Also, I think the "not planning a new release soon" is WRT consumer GPUs and not data center GPUs (MI300). :)

Regards,
SB
True dat on the software, here's hoping it improves! I haven't checked too deep, but is AMD's software open source or proprietary? If it's open source it could catch up to nVidia in a hurry, if not then we'll have more great AMD hardware with bad software...where have I heard that before?
 
Seems AMD is not planning a new release soon, they are not going to announce anything consumer GPU related at Computex, and they already released N33 before N32, which is unheard of. One of the comments on the matter caught my attention.

The rumor was that RDNA 4 would be ready for the end of the year. If that is still the case it may be best for them to limit rdna 3 releases.
 
 
Status
Not open for further replies.
Back
Top